Older people are at increased risk of many adverse health outcomes, including dementia and depression, that burden the global health system. This paper presents algorithms for the large-scale assessment of daily walking speeds. We hypothesize that (i) data from wrist-worn sensors can be used to assess walking speed accurately; and that (ii) maximal daily walking speed is a better predictor of health outcomes than usual daily walking speed. First, algorithms were developed and tested using data from 101 participants aged 19 to 91 (47 ± 18) years. Participants wore an AX3 accelerometer (Axivity, UK) on their dominant wrist while undertaking daily life activities with electronic walkway data used for ground truth. Subsequently, prediction models for dementia, depression and death were developed using the data of 47,406 participants (≥ 60 years) from the UK Biobank study. Daily walking speeds were derived from 7-day AX3 data with time-to-events using electronic health records. The accuracy of derived walking speeds was assessed using root mean square error (RMSE). Time-to-events were modelled using Cox regression with inverse hazard ratios reported for univariable models and Harrell's concordance for multivariable models. Derived walking speeds had an RMSE of between 3% and 4% depending on arm position. We found that for simple models, maximal walking speed was significantly better than usual walking speed at predicting time to dementia (1.62 vs 1.34), depression (1.29 vs 1.17) and death (1.56 vs 1.27). However, the addition of known risk factors in subsequent multivariable models reduced the apparent benefit of using maximal as opposed to usual daily walking speed as the gait parameter. In summary, walking speed was accurately measured with a wrist-worn device, and maximal daily waking speed may be better than usual daily walking speed at predicting some adverse health outcomes.Clinical Relevance- This study demonstrated the validity of using a simple and unobtrusive wrist-worn sensor to remotely assess daily walking speed. As a single, modifiable and easily understood measure, maximal walking speed was shown to be better than usual walking speed at predicting time-to-dementia, depression and death. Therefore, the inclusion of maximal daily walking speed into screening programs and clinical interventions presents a promising area for further research.